Fix polygon area calculation: use Shoelace formula #2927

merkispavel · 2021-03-09T05:57:18Z

https://en.wikipedia.org/wiki/Shoelace_formula

Issue

Isochrone api creates polygon away from the origin location. The reason is the broken polygon_area function. The area of the small polygon turnes out to be larger than the whole convex-hull-ish polygon. So if you set denoise = 1 you get the issue itself.

If you want to reproduce

request: '{"locations":[{"lat":51.2491202,"lon":0}],"costing":"pedestrian","contours":[{"time":20}], "show_locations":true, "polygons":true, "denoise":1}'

planet.tar.zip

Tasklist

Add tests
Add #fixes with the issue number that this PR addresses
Update the docs with any new request parameters or changes to behavior described
Update the changelog

Requirements / Relations

Link any requirements here. Other pull requests this PR is based on?

https://en.wikipedia.org/wiki/Shoelace_formula

merkispavel · 2021-03-09T07:37:56Z

I noticed that inner polygons have positive area sometimes. I think this might be the cause of the failed tests. Anyway, investigating..

kevinkreiser · 2021-03-09T14:27:17Z

@merkispavel just an FYI recently @nilsnolde was also computing polygon area (using boost) in #2750 i wonder if both PRs could use the same method of computing polygon area. just wanted to flag it for you to have a look

dnesbitt61 · 2021-03-09T15:06:42Z

Polygon area (line 552 in midgard/util.cc) I believe should be:

area += (p1->first * p2->second) - (p1->second * p2->first)

This should work for any simple polygon.

dnesbitt61 · 2021-03-09T16:06:17Z

I verified that the change described above is correct. I wrote a simple test with a rectangle and the current polygon_area computation fails and the update is correct. Problem is this breaks the isochrone tests.

dnesbitt61 · 2021-03-09T16:18:33Z

I think the standard convention is for counter-clockwise orientations to produce positive area and clockwise orientations should yield negative area. A simple test (rectangle) could be added to test/util_midgard.cc:

TEST(UtilMidgard, Area) {
  // Clockwise polygon (rectangle). Should return negative area.
  using poly_t = std::vector<Point2>;
  poly_t poly{ {0, 0}, {0, 20}, {20, 20}, {20, 0}, {0, 0} };
  auto area = polygon_area(poly);
  EXPECT_EQ(area, -400.0);

  // Reverse the polygon (ccw) - should be positive area
  std::reverse(std::begin(poly), std::end(poly));
  area = polygon_area(poly);
  EXPECT_EQ(area, 400.0);
}

merkispavel · 2021-03-09T20:13:17Z

Polygon area (line 552 in midgard/util.cc) I believe should be:

area += (p1->first * p2->second) - (p1->second * p2->first)

This should work for any simple polygon.

I verified that the change described above is correct. I wrote a simple test with a rectangle and the current polygon_area computation fails and the update is correct. Problem is this breaks the isochrone tests.

Right, these formulas are equivalent. I was too lazy to edit more than 2 symbols :) sorry

merkispavel · 2021-03-09T20:17:01Z

I think the standard convention is for counter-clockwise orientations to produce positive area and clockwise orientations should yield negative area. A simple test (rectangle) could be added to test/util_midgard.cc:
TEST(UtilMidgard, Area) {
  // Clockwise polygon (rectangle). Should return negative area.
  using poly_t = std::vector<Point2>;
  poly_t poly{ {0, 0}, {0, 20}, {20, 20}, {20, 0}, {0, 0} };
  auto area = polygon_area(poly);
  EXPECT_EQ(area, -400.0);

  // Reverse the polygon (ccw) - should be positive area
  std::reverse(std::begin(poly), std::end(poly));
  area = polygon_area(poly);
  EXPECT_EQ(area, 400.0);
}

Totally agreed! I have more changes locally with unittests + comments, haven't pushed it yet
Moreover, some changes are needed not to lose polygon orientation in the isochrone code. This is done as well(haven't pushed it yet). I just need to figure out whether to push it to this branch or split to ease its review.
Anyway, I'll ping you when its ready(probably tomorrow)

merkispavel · 2021-03-11T12:16:05Z

Another PR should be merged first in order to fix tests #2932

merkispavel · 2021-03-11T12:17:10Z

src/midgard/util.cc

  for (auto p1 = polygon.cbegin(), p2 = std::next(polygon.cbegin()); p2 != polygon.cend();
       ++p1, ++p2) {
-    area += (p1->first + p2->first) * (p1->second + p2->second);
+    area += p1->first * p2->second - p1->second * p2->first;


@dnesbitt61 I changed the formula to its classic form not to confuse people in the future

dnesbitt61

Looks good, thanks!

kevinkreiser · 2021-04-06T13:08:36Z

@merkispavel if you could visually inspect the differences in the isochrone unit test and then update the values then we can get CI passing and we can merge it. looks good to me

* Track segment orientation This is need to correctly join segments into rings * Update changelog * Remove ring reversing * Fix: use multiline comment

…rmula # Conflicts: # test/isochrone.cc

merkispavel · 2021-04-07T13:00:35Z

@kevinkreiser @dnesbitt61 the tests are fixed. Looks shippable in my view

kevinkreiser · 2021-04-07T14:25:44Z

CHANGELOG.md

+   * FIXED: Skip bindings if there's no Python development version [#2893](https://github.com/valhalla/valhalla/pull/2893)
   * FIXED: Skip bindings if there's no Python development version [#2893](https://github.com/valhalla/valhalla/pull/2878)


this seems wrong 😄

The wrong link was used before so automerge decided to keep both entries. Fixed, thanks

kevinkreiser · 2021-04-07T14:29:05Z

valhalla/midgard/gridded_data.h

@@ -98,7 +98,7 @@ template <std::size_t dimensions_t> class GriddedData : public Tiles<PointLL> {
    typename PointLL::first_type s[5]; // Values at the tile corners and center


why is the code from #2932 in this PR as well? something seems messed up here?

Originally i found 2 issues at once(polygon area bug and orientation bug) so I decided to make a hierarchy master <- 2927 <- 2932 for some reason 🤦(probably because of the unit tests) Now I think It brought more mess than clearness. Anyway, I had to fix unit tests so it didn't bring any profit
Would you like to split them @kevinkreiser ? (I guess I do, it'll just take a little more time and a few more approves)

oh no, no problem at all, i just missed a very important detail:

the previous one was merged into this branch not master. i completely missed that. this is fine!

kevinkreiser · 2021-04-07T14:47:46Z

valhalla/midgard/gridded_data.h

    std::array<std::function<void()>, 10> cases{
-        []() {},
+        [&]() { from_pt = to_pt = {}; },


do you want to change that back to a no-op?

kevinkreiser

LGTM, previous review was the difficult part. fixing the area calculation was straight forward

merkispavel added 2 commits March 9, 2021 08:30

Fix polygon_area calculation: use Shoelace formula

6dd02d4

https://en.wikipedia.org/wiki/Shoelace_formula

Update CHANGELOG.md

0b2c717

merkispavel added 3 commits March 11, 2021 13:05

Use classic form of polygon area formula

afce0fc

Add PolygonArea test

6ff8c0e

Merge branch 'master' into fix-polygon-area-formula

e4b33b3

merkispavel commented Mar 11, 2021

View reviewed changes

merkispavel requested a review from dnesbitt61 April 6, 2021 10:39

dnesbitt61 reviewed Apr 6, 2021

View reviewed changes

merkispavel added 3 commits April 6, 2021 23:27

Orient segments/rings according to right-hand rule (#2932)

a5e6388

* Track segment orientation This is need to correctly join segments into rings * Update changelog * Remove ring reversing * Fix: use multiline comment

Merge remote-tracking branch 'origin/master' into fix-polygon-area-fo…

3d7ac7b

…rmula # Conflicts: # test/isochrone.cc

Fix tests after master merge

6ce81b3

merkispavel requested a review from dnesbitt61 April 7, 2021 13:00

kevinkreiser reviewed Apr 7, 2021

View reviewed changes

Update CHANGELOG.md

b51a0ae

merkispavel requested a review from kevinkreiser April 7, 2021 14:27

kevinkreiser reviewed Apr 7, 2021

View reviewed changes

kevinkreiser previously approved these changes Apr 7, 2021

View reviewed changes

Replace unused case with no-op

49d2600

merkispavel dismissed kevinkreiser’s stale review via 49d2600 April 7, 2021 14:51

kevinkreiser previously approved these changes Apr 7, 2021

View reviewed changes

Merge branch 'master' into fix-polygon-area-formula

c2e7871

merkispavel dismissed kevinkreiser’s stale review via c2e7871 April 7, 2021 15:24

purew approved these changes Apr 7, 2021

View reviewed changes

merkispavel merged commit 2e74c32 into master Apr 7, 2021

purew deleted the fix-polygon-area-formula branch April 7, 2021 16:25

dnesbitt61 mentioned this pull request Apr 16, 2021

Performance Drop in Long Bicycle Route #3007

Closed

nilsnolde mentioned this pull request Jul 19, 2021

Remove "bg::correct" again #3217

Closed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Fix polygon area calculation: use Shoelace formula #2927

Fix polygon area calculation: use Shoelace formula #2927

merkispavel commented Mar 9, 2021 •

edited

Loading

merkispavel commented Mar 9, 2021

kevinkreiser commented Mar 9, 2021

dnesbitt61 commented Mar 9, 2021

dnesbitt61 commented Mar 9, 2021

dnesbitt61 commented Mar 9, 2021

merkispavel commented Mar 9, 2021

merkispavel commented Mar 9, 2021

merkispavel commented Mar 11, 2021

merkispavel Mar 11, 2021

dnesbitt61 Mar 11, 2021

dnesbitt61 left a comment

kevinkreiser commented Apr 6, 2021

merkispavel commented Apr 7, 2021

kevinkreiser Apr 7, 2021

merkispavel Apr 7, 2021

merkispavel Apr 7, 2021

kevinkreiser Apr 7, 2021

merkispavel Apr 7, 2021

kevinkreiser Apr 7, 2021

kevinkreiser Apr 7, 2021

kevinkreiser left a comment

		* FIXED: Skip bindings if there's no Python development version [#2893](https://github.com/valhalla/valhalla/pull/2893)
		* FIXED: Skip bindings if there's no Python development version [#2893](https://github.com/valhalla/valhalla/pull/2878)

		@@ -98,7 +98,7 @@ template <std::size_t dimensions_t> class GriddedData : public Tiles<PointLL> {
		typename PointLL::first_type s[5]; // Values at the tile corners and center

Fix polygon area calculation: use Shoelace formula #2927

Fix polygon area calculation: use Shoelace formula #2927

Conversation

merkispavel commented Mar 9, 2021 • edited Loading

Issue

If you want to reproduce

Tasklist

Requirements / Relations

merkispavel commented Mar 9, 2021

kevinkreiser commented Mar 9, 2021

dnesbitt61 commented Mar 9, 2021

dnesbitt61 commented Mar 9, 2021

dnesbitt61 commented Mar 9, 2021

merkispavel commented Mar 9, 2021

merkispavel commented Mar 9, 2021

merkispavel commented Mar 11, 2021

Choose a reason for hiding this comment

Choose a reason for hiding this comment

dnesbitt61 left a comment

Choose a reason for hiding this comment

kevinkreiser commented Apr 6, 2021

merkispavel commented Apr 7, 2021

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

kevinkreiser left a comment

Choose a reason for hiding this comment

merkispavel commented Mar 9, 2021 •

edited

Loading